Comparative RNA-Sequencing Analysis Benefits a Pediatric Patient With Relapsed Cancer.

Clinical detection of sequence and structural variants in known cancer genes points to viable treatment options for a minority of children with cancer.1 To increase the number of children who benefit from genomic profiling, gene expression information must be considered alongside mutations.2,3 Although high expression has been used to nominate drug targets for pediatric cancers,4,5 its utility has not been evaluated in a systematic way.6 We describe a child with a rare sarcoma that was profiled with whole-genome and RNA sequencing (RNA-Seq) techniques. Although the tumor did not harbor DNA mutations targetable by available therapies, incorporation of gene expression information derived from RNA-Seq analysis led to a therapy that produced a significant clinical response. We use this case to describe a framework for inclusion of gene expression into the clinical genomic evaluation of pediatric tumors.

Clinical detection of sequence and structural variants in known cancer genes points to viable treatment options for a minority of children with cancer.[1] To increase the number of children who benefit from genomic profiling, gene expression information must be considered alongside mutations.[2,3] Although high expression has been used to nominate drug targets for pediatric cancers,[4,5] its utility has not been evaluated in a systematic way.[6] We describe a child with a rare sarcoma that was profiled with whole-genome and RNA sequencing (RNA-Seq) techniques. Although the tumor did not harbor DNA mutations targetable by available therapies, incorporation of gene expression information derived from RNA-Seq analysis led to a therapy that produced a significant clinical response. We use this case to describe a framework for inclusion of gene expression into the clinical genomic evaluation of pediatric tumors.

CASE SUMMARY

Patient 1 was diagnosed at 8 years of age with a left tentorial-based CNS sarcoma after a 2-week history of nausea, lethargy, and diplopia. Clinical workup confirmed that the tumor was primary to the brain (Figs 1A and 1B). Histology revealed a mitotically active, epithelioid-to-spindled cell tumor in patternless sheets, interrupted by thick fibrous bands and foci of necrosis (Figs 1C to 1D). Immunohistochemistry revealed diffuse positivity for vimentin, desmin, neuron-specific enolase, epithelial membrane antigen, and CD99 (Figs 1E to 1H). Focal immunohistochemical positivity was observed for pan-cytokeratin (AE1/AE3) and synaptophysin. The tumor was negative for glial fibrillary acidic protein (GFAP), Wilms tumor 1 (WT1), myo-D1, myogenin, smooth muscle actin, nonphosphorylated and phosphorylated neurofilament protein, CD34, CD31, HMB-45, S-100, leukocyte common antigen, and BAF47/INI-1 (retained nuclear positivity). The Ki67 proliferative index was 9%. A diagnosis of desmoplastic small round cell tumor (DSRCT) was favored initially.[7] Because EWSR1 breakapart fluorescence in situ hybridization confirmed an EWSR rearrangement but concomitant WT1 breakapart fluorescence in situ hybridization was negative, the molecular criterion for DSRCT was not met, and a final diagnosis of poorly differentiated sarcoma, not otherwise specified, was rendered. The patient received six cycles of induction chemotherapy—ifosfamide, carboplatin, and etoposide—followed by autologous stem-cell transplantation with a high-dose preparative regimen of carboplatin, thiotepa, and etoposide as well as 54 Gy of focal radiation to the location of the original tumor. After a 2-year remission, the tumor recurred with numerous pulmonary lesions in all lobes. The histologic characteristics of the metastasis were identical to the primary tumor. The patient enrolled in the Personalized OncoGenomics (POG)[3] study, which offers whole-genome sequencing (WGS) and transcriptome sequencing and analysis to identify drivers and potential therapeutic options of relapsed solid tumors for children and adults in British Columbia.

Fig 1.

Case clinical information. (A) Preoperative magnetic resonance imaging (MRI). T2-weighted coronal sequence revealed a large, primarily hyperintense tumor that arises from the ventral aspect of the left tentorium, invaginating into the superior aspect of the cerebellum and causing diffuse edema therein. (B) Preoperative MRI. T1-weighted axial sequence with gadolinium revealed strong enhancement. (C) Routine tumor histology. Hematoxylin and eosin (H&E)–stained representative section revealed a combination of epithelioid and spindled tumor cells among thick fibrous bands (× 200 magnification). (D) Routine tumor histology. Higher magnification depicts epithelioid tumor cells (eg, long arrow) and a mitotic figure (short arrow); many of the tumor cells exhibit somewhat vacuolated cytoplasm (H&E; × 400 magnification). (E-H) Tumor immunohistochemistry. Strong diffuse cytoplasmic immunostaining is exhibited for (E) desmin and (F) neuron-specific enolase (NSE). Diffuse membranous immunostaining is appreciated for (G) epithelial membrane antigen (EMA), and (H) CD99 (all photomicrographs taken at × 200 magnification).

Biopsy material from a lung metastasis was characterized with WGS and RNA-Seq, and peripheral blood was characterized with WGS.[3] The analysis of the sequencing data revealed an EWSR1-ATF1 gene fusion (Appendix Fig A1, online only); although this finding is most suggestive of a clear cell sarcoma, no immunohistochemical support for this diagnosis could be established.[8,9] The POG team identified three somatic variants of unclear therapeutic significance: PDGFRA p.V299F, PRKCB p.D341N and SVIL p.L1374R. No germline single-nucleotide variants with established cancer relevance were detected.[10] Although no therapy was available to target the EWSR1-ATF1 fusion protein directly, the POG RNA-Seq–derived gene expression analysis identified high expression of downstream genes IL6 and JAK1. The finding of the JAK1 overexpression was corroborated by comparative RNA-Seq analysis at the University of California Santa Cruz.

Fig A1.

Schematic representation of the EWSR1-ATF1 fusion. Two forms of the fusion were identified; the EWSR1-ATF1 form had higher expression than the reciprocal ATF1-EWSR1 form. BZIP, basic leucine zipper domain, which mediates sequence specific DNA binding properties and the leucine zipper that is required to hold together (dimerize) two DNA binding regions; pKID, phosphorylated kinase-inducible-domain; RRM, RNA recognition motif; ZF, Zn-finger in Ran binding protein and others.

COMPARATIVE RNA-SEQ ANALYSIS

In accordance with the US Food and Drug Administration guidelines,[6] we focused on relative rather than absolute gene expression levels and sought to develop a framework for the analysis of RNA-Seq data from patients. We compared the RNA-Seq–derived tumor gene expression profile of patient 1 with similarly derived profiles of 10,668 samples that represented 38 pediatric and adult tumor types studied by the The Cancer Genome Atlas (TCGA) and Therapeutically Actionable Research to Generate Effective Treatments (TARGET).[11,12] RNA-Seq reads from different laboratories were reanalyzed with a single computational pipeline to reduce batch effects.[13] We searched for tumors in this homogeneously processed compendium in which expression profiles were similar to those of patient 1 by using TumorMap.[14] The tumor gene expression profile for patient 1 resembled lung cancers (Fig 2A), the site of the metastasis. The metastatic sample contained 76% tumor cells, estimated by a POG pathologist, which suggested that most of the gene expression information came from the tumor cells. Lung adenocarcinoma (LUAD) samples formed four groups in the TumorMap (Fig 2B) and the sarcoma of patient 1 clustered with the 354 LUAD tumors of the terminal respiratory unit and proximal-inflammatory molecular subtype (Fig 2C), associated with the activation of receptor tyrosine kinases (RTK).[15] To define the transcriptional programs that drove placement of the patient’s tumor with the lung cancers, we conducted Gene Set Enrichment Analysis[16] with genes differentially expressed between the LUAD cluster that contained the tumor of patient 1 (n = 354) and the remaining samples in the compendium (n = 10,314); we also repeated this analysis and compared the cluster for patient 1 with the remaining LUAD samples (n = 529). Both analyses revealed the overexpression of members of the IL6/JAK/STAT3 signaling pathway (Appendix Fig A2), which suggests that the activation of shared signaling programs likely contributed to the tumor transcriptional phenotype of patient 1 in addition to the site of the metastatic sample. We next searched for genes that were significantly overexpressed in the patient’s tumor compared with the whole compendium and compared with only the sarcomas by using outlier statistics[3,17] (Data Supplement). To explicitly subtract the effect of the lung cell expression, we also searched for outlier genes compared with 529 LUAD tumors; 787 genes, including druggable targets JAK1, ALK, NTRK1, and CCND1, emerged as overexpression outliers in all analyses (Data Supplement).

Fig 2.

RNA-Seq–based gene expression profile for patient 1 visualized in the context of the reference cohort of 38 adult and pediatric tumor types. (A) A projection of the entire tumor cohort in two dimensions according to the TumorMap method. Individual tumors are represented by hexagons, and colored tumors by the tumor type, as indicated in the graphic. The tumor in patient 1 is shown in green within the lung tumors. (B) Lung adenocarcinoma (LUAD) tumors are found in four main regions of the TumorMap visualization. LUAD tumors are depicted in orange, whereas all other tumor types are in gray. (C) A zoomed-in view of the cluster for patient 1 and the surrounding area that contains LUAD tumors, now colored according to LUAD molecular subtypes (proximal proliferative, blue; proximal inflammatory, green; terminal repiratory unit, red). Unclassified samples are colored in gray.

Fig A2.

Relative expression levels of EWSR1, ATF1, and JAK1, compared to different cohorts of adult (TCGA) and pediatric (TARGET) tumours. Patient 1 tumor gene expression levels are indicated by the red vertical line. The two cohorts used for the outlier analysis are highlighted in gold. Ped, pediatric tumor types; TARGET, Therapeutically Actionable Research to Generate Effective Treatments; TGCA, The Cancer Genome Atlas.

MOLECULAR RATIONALE FOR CLINICAL DECISION MAKING

We speculated that the activation of RTKs contributed to JAK over-expression in patient 1’s tumor.[18,19] Increased expression of ATF1 and its transcriptional targets, TOP2A, CALCA, and IL6, was observed, presumably as a result of constitutive transcriptional activation by EWSR1-ATF1 (Fig 3A; Appendix Fig A3). ATF1 can activate the transcription of JAK1,[8] providing another potential mechanism for the observed high expression of the IL6/JAK/STAT3 pathway. Consolidating the fusion-based and the RTK-based mechanisms of IL6/JAK/STAT3 activation, we reconstructed a candidate pathway driving patient 1’s cancer (Fig 3, Appendix Fig A4). The POG molecular tumor board suggested targeting either JAK (with ruxolitinib) or ALK (with crizotinib). A decision was made to use ruxolitinib given (1) the over-expression of ATF1 target genes, (2) the over-expression of JAK1, and (3) the available pediatric dosing information.[20] In addition, ruxolitinib was favored over crizotinib because it targets downstream of both EWSR1-ATF1 and the over-expressed RTKs, whereas crizotinib only targets the RTKs. We recognize that a combination therapy targeting both ALK and JAK may have been appropriate on the basis of the molecular findings. However, we were unable to use drug combinations that have not been through phase 1 testing, highlighting the need for more pediatric combination therapy trials.

Fig 3.

Molecular rationale for using ruxolitinib to treat the sarcoma in patient 1. (A) Candidate pathway that drove tumorigenesis in patient 1 was reconstructed on the basis of outlier analysis, differential expression analysis compared with normal tissues, copy number information, and literature mining (Appendix Methods). Both EWSR1-ATF1 and receptor tyrosine kinases NTRK1 and ALK can contribute to the activation of IL6/JAK/STAT3 signaling. All gene expression outliers depicted in the figure (gene names written in red font) were significant in all three comparisons: patient 1 versus all cancers, patient 1 versus lung adenocarcinomas, and patient 1 versus sarcomas. JAK1, the molecular target of ruxolitinib, is indicated with a yellow lightning bolt. (B) The tumor in patient 1 expresses JAK1 at a strikingly higher level than those seen in all 10,668 tumors, which are represented by 38 tumor types studied by the TCGA[11] and TARGET (denoted PANCAN),[12] including lung adenocarcinomas (LUADs) and sarcomas (SARCs). (C) JAK1 is an attractive molecular target for patient 1’s tumor because it is downstream of the EWSR1-ATF1 fusion and the activate receptor tyrosine kinases (RTKs). It was also identified as over expressed by gene expression outlier analysis.

Fig A3.

Gene Set Enrichment Analysis (GSEA) of the gene scores obtained through differential gene expression analysis comparing patient 1’s TumorMap cluster to the whole reference compendium (left) and to the remaining TCGA LUAD samples (right) identifies IL6/JAK/STAT3 pathway as one of the most significantly enriched pathways in genes differentially overexpressed in patient 1’s TumorMap cluster.

Fig A4.

Full candidate pathway that represents molecular drivers of tumorigenesis in the sarcoma of patient 1. We reconstructed this pathway on the basis of the outlier analysis, differential gene expression analysis, copy number information and literature mining (Appendix Methods). A simplified version of this pathway is presented in Figure 3A. Both EWSR1-ATF1 and receptor tyrosine kinases PDGFRB, NTRK1, ALK, and FGFR1 can contribute to the activation of IL6/JAK/STAT3 signaling. All gene expression outliers depicted in this figure were significant in all three comparisons: patient 1 versus all cancers, patient 1 versus lung adenocarcinomas, and patient 1 versus sarcomas. JAK1, the molecular target of ruxolitinib, is indicated with a yellow lightning bolt. TCGA, The Cancer Genome Atlas.

CLINICAL RESPONSE

At the initiation of therapy with single-agent ruxolitinib, patient 1 had severe nausea and lethargy, was mostly bed-ridden, and had a Lansky play-performance score[21] of 60 (Fig 4). Within a week of ruxolitinib initiation, his mother reported a dramatic improvement in his energy level and complete resolution of his nausea. The patient tolerated this therapy without significant toxicity and had stabilization of the previously rapidly growing lung nodules by RECIST,[22] and his Lansky score improved to 90 to 100 for 5 months. The patient then exhibited a sudden enlargement in one lung lesion, detected during routine imaging, although most of the other lesions remained stable. Ruxolitinib was discontinued, and focal palliative radiation was administered to the one lesion in the left lower lobe for pain control. Within 2 months of ruxolitinib discontinuation, the symptoms of nausea, extreme fatigue, and weight loss returned, and the lung lesions progressed. The family requested that ruxolitinib be restarted for quality of life, and the patient again showed dramatic improvement in clinical status and an unexpected prolonged period of stable disease until dose reduction because of myelosuppression was required. After the dose reduction, rapid progression of the pulmonary lesions resulted in death 23 months after the original relapse.

Fig 4.

Clinical response of patient 1 to ruxolitinib (Ruxi). Ruxi was initiated at day 0 at 40 mg twice per day. The patient’s body weight was 40.3 kg, and the Lansky play-performance score was 60 at the time of treatment initiation. During treatment, the patient’s status improved significantly according to both body weight and Lansky play-performance score, which reached normal levels; thus, the Ruxi dose was increased to 60 mg twice per day at day 29. The patient continued to receive this dose until disease progression according to computed tomography occurred at day 188 (Ruxi stopped; progression). Ruxi was restarted at day 348, and a response again was noted according to both body weight and Lansky performance score.

DISCUSSION

To our knowledge, this is the first report of a pediatric patient with cancer who benefited from cross-tumor gene expression comparisons. Cross-tumor analyses have been used in the TCGA[11] and POG studies[3,23,24]; however, a computational framework is necessary for their clinical implementation. This case is also, to our knowledge, the first report of use of a JAK inhibitor to treat a sarcoma. Previous functional studies implicated STAT3 as an oncogene in sarcomas[25]; the current case report builds on this work and prompts investigation into the potential clinical utility of targeting this pathway. Of note was the patient’s marked and rapid clinical response to treatment, which suggests that response may have been related to the modulation of cytokine expression by the medication. Although the clinical benefit of ruxolitinib was apparent, it was challenging to quantify. Ultimately, a randomized clinical trial is necessary to assess the benefit of molecular approaches compared with the standard of care.

The case also highlights tumor heterogeneity: although the majority of the metastases remained stable, one lesion rapidly became resistant. Despite documented progressive disease, this case benefited from resumption of the medication: the patient’s clinical status and many metastatic lesions were responsive to retreatment. A serial molecular analysis of the heterogeneous lesions could inform the mechanisms of resistance; however, it was not pursued because of the family’s wishes. To characterize the intratumor heterogeneity of therapeutic response, we consider follow-up biopsies, and those decisions are weighed against the risks to the patient.

Table A1.

Antibodies and Vendors